For years, the answer to “can I use conda for R?” was “only if your packages happen to be on conda-forge”. That’s no longer the case: with conda-forge, bioconda, and packages built from source, most of CRAN and Bioconductor is usable in a Pixi project. This article has the numbers, for deciding whether Pixi is ready for your R work.
The numbers are from 12 October 2026, with Pixi 0.81.0. They’ll change as conda-forge, bioconda and Pixi do.
CRAN
Of CRAN’s 25,299 packages:
| Packages | Share | |
|---|---|---|
| On conda-forge | 3,423 | 14% |
| Built from source by rpix | 16,751 | 66% |
| Usable through Pixi and rpix | 20,174 | 80% |
| Not yet: a dependency isn’t on conda-forge | 4,754 | 19% |
| Not yet: needs a system library R doesn’t bring | 371 | 1% |
conda-forge has fewer packages than CRAN, but the ones that matter most: 945 of the 1,000 packages other CRAN packages depend on most, and 93% of its R packages are at CRAN’s current version.
Built from source is what pixi_add() offers for a CRAN package that isn’t on conda-forge: Pixi builds it from CRAN’s source, with its dependencies from conda-forge (see Finding packages). A package counts when all its dependencies (Depends, Imports and LinkingTo) are on conda-forge or come with R, and its SystemRequirements need nothing beyond what conda-forge’s R brings (such as curl, cairo, libxml2 or GSL) and the build tools Pixi and rpix add (Rust, CMake, pkg-config, Java). Building one takes a minute or two. A random sample of 25 of them all built and loaded, on macOS.
What’s still missing
Most of the packages that don’t work yet depend on a package that isn’t on conda-forge either. For 3,938 of these 4,754, that dependency can be built from source too, so once Pixi can build packages that depend on other packages built from source (prefix-dev/pixi#6925), rpix can build the whole chain. Most of the rest need a Bioconductor package (often limma, Biostrings, GenomicRanges or S4Vectors), which needs prefix-dev/pixi#6994, or a system library such as GDAL or libsndfile.
| Usable | Share | |
|---|---|---|
| Today | 20,174 | 80% |
| Building missing dependencies from source too | 24,112 | 95% |
| And Bioconductor dependencies from bioconda | 24,592 | 97% |
Bioconductor
bioconda has essentially all of Bioconductor 3.22, for R 4.5: 2,291 of its 2,300 software packages, 98% of them at their 3.22 version or newer. Add them with a bioc:: prefix, or without one: pixi_add() looks on bioconda when a package isn’t on conda-forge.
Some things to know:
- Use R 4.5. Bioconductor 3.23, for R 4.6, isn’t on bioconda yet. That waits on conda-forge building its R packages for R 4.6, which only a few are so far.
-
Annotation and data packages, such as org.Hs.eg.db, download their data with an install script, which Pixi only runs in projects that allow it.
pixi_add()offers to allow it. - Windows: bioconda doesn’t build its packages for Windows. Bioconductor packages that don’t depend on other Bioconductor packages do build from source there, from Bioconductor’s git mirror on GitHub, but rpix doesn’t offer that yet.
Platforms
The source builds were measured on macOS. Packages of each kind (pure R, R’s recommended packages, C, C++, Fortran, Rust and Bioconductor) also build on Linux, and on Windows with what rpix adds for building there.
How these were counted
-
CRAN: every package in CRAN’s package database, with its dependencies and
SystemRequirements. -
conda-forge and bioconda: the packages in their channel data, for osx-64, matched by name (
r-<name>andbioconductor-<name>). - System requirements: read with Posit’s r-system-requirements rules. A language standard (“C++17”), make, and tools only needed when the package runs (pandoc, Python) count as needing nothing.
The script that counts them is data-raw/package-availability.R.